Consider the following two "types" of interpolation. In one case, our model passes through all observed data, in the other one, it doesn't. Do these types of interpolations have a name?
If I recall correctly:
- Some authors would say that in (a) the model "fully interpolates the data"
- In classification, I believe some authors may say that model (a) "fully shatters the data".
- I think it's also common to say that model (a) "memorizes" the data.
I know that (a) is fitting the data more closely (it has a higher risk of overfitting), but I am wondering if there is a name or denomination for the actual phenomenon of having a model that takes on the actual values of the observed data instead of "averaging" through it.